Rs in the Results section. Our maps are designed to facilitate the identification of spatial clusters of sites that are extreme outliers of the same type in the same day. For each day there are two series of maps, one for extreme CCX282-B price positive outliers, another for extreme negative outliers. Each has six panels grouped in three rows (sites with unusual call frequency, sites with unusual movement frequency andPLOS ONE | DOI:10.1371/journal.pone.0120449 March 25,10 /Spatiotemporal Detection of Unusual Human Population BehaviorFig 6. Daily call volume time series of probabilities associated with site 361. The means of the estimated probabilities of making more calls are shown in blue. The gray band gives the intervals between the 2.5 and the 97.5 quantiles of the estimated probabilities. The dots indicate which days in this time series are extreme positive outliers (red) and extreme negative outliers (green). The red (green) squares indicate the confidence probabilities that a day is a positive (negative) extreme outlier. The confidence probabilities are shown only for the days that have been classified as an outlier at least once. doi:10.1371/journal.pone.0120449.gFig 7. Daily movement frequency time series of probabilities associated with site 361. The means of the estimated probabilities of being more mobile are shown in blue. The gray band gives the intervals between the 2.5 and the 97.5 quantiles of the estimated probabilities. The dots indicate which days in this time series are extreme positive outliers (red) and extreme negative outliers (green). The red (green) squares indicate the confidence probabilities that a day is a positive (negative) extreme outlier. The confidence probabilities are shown only for the days that have been classified as an outlier at least once. doi:10.1371/journal.pone.0120449.gPLOS ONE | DOI:10.1371/journal.pone.0120449 March 25,11 /Spatiotemporal Detection of Unusual Human Population Behaviorsites with both). The maps in the first column show the locations of the sites with disturbances and the locations of the sites without disturbances. Sites that are extreme outliers of the same type, positive or negative, and are on average spatially closer to each other than to other sites, comprise a cluster. We suggest that the anomalous behavior in a spatial cluster of sites is most likely caused by a single event. Separate spatial clusters are more likely to represent different events in different places. The two maps in the third row of Fig. 2 show 10 sites with unusually high call volume and movement frequency. Nine of these sites are all the sites located in Rwanda within a 50 km radius from epicenters of the Lake Kivu earthquakes, thus it is very likely that this unusual behavioral pattern was caused by the earthquakes. The maps in the first row of Fig. 3 show two sites with unusually low call volume, and the maps on the second and third row show that one of these sites also recorded unusually low movement frequency. Despite these unusual behavioral patterns occurring on the same day as the Lake Kivu earthquakes, they were likely caused by another event of a different type, not only because they led to lower rather than higher than usual call and movement frequency, but also because they occurred far from the epicenters of the earthquakes. Instead of simple order ICG-001 visual examination of the maps, we use a systematic method to identify spatial clusters of sites. Our goal is to place sites together in a cluster if they.Rs in the Results section. Our maps are designed to facilitate the identification of spatial clusters of sites that are extreme outliers of the same type in the same day. For each day there are two series of maps, one for extreme positive outliers, another for extreme negative outliers. Each has six panels grouped in three rows (sites with unusual call frequency, sites with unusual movement frequency andPLOS ONE | DOI:10.1371/journal.pone.0120449 March 25,10 /Spatiotemporal Detection of Unusual Human Population BehaviorFig 6. Daily call volume time series of probabilities associated with site 361. The means of the estimated probabilities of making more calls are shown in blue. The gray band gives the intervals between the 2.5 and the 97.5 quantiles of the estimated probabilities. The dots indicate which days in this time series are extreme positive outliers (red) and extreme negative outliers (green). The red (green) squares indicate the confidence probabilities that a day is a positive (negative) extreme outlier. The confidence probabilities are shown only for the days that have been classified as an outlier at least once. doi:10.1371/journal.pone.0120449.gFig 7. Daily movement frequency time series of probabilities associated with site 361. The means of the estimated probabilities of being more mobile are shown in blue. The gray band gives the intervals between the 2.5 and the 97.5 quantiles of the estimated probabilities. The dots indicate which days in this time series are extreme positive outliers (red) and extreme negative outliers (green). The red (green) squares indicate the confidence probabilities that a day is a positive (negative) extreme outlier. The confidence probabilities are shown only for the days that have been classified as an outlier at least once. doi:10.1371/journal.pone.0120449.gPLOS ONE | DOI:10.1371/journal.pone.0120449 March 25,11 /Spatiotemporal Detection of Unusual Human Population Behaviorsites with both). The maps in the first column show the locations of the sites with disturbances and the locations of the sites without disturbances. Sites that are extreme outliers of the same type, positive or negative, and are on average spatially closer to each other than to other sites, comprise a cluster. We suggest that the anomalous behavior in a spatial cluster of sites is most likely caused by a single event. Separate spatial clusters are more likely to represent different events in different places. The two maps in the third row of Fig. 2 show 10 sites with unusually high call volume and movement frequency. Nine of these sites are all the sites located in Rwanda within a 50 km radius from epicenters of the Lake Kivu earthquakes, thus it is very likely that this unusual behavioral pattern was caused by the earthquakes. The maps in the first row of Fig. 3 show two sites with unusually low call volume, and the maps on the second and third row show that one of these sites also recorded unusually low movement frequency. Despite these unusual behavioral patterns occurring on the same day as the Lake Kivu earthquakes, they were likely caused by another event of a different type, not only because they led to lower rather than higher than usual call and movement frequency, but also because they occurred far from the epicenters of the earthquakes. Instead of simple visual examination of the maps, we use a systematic method to identify spatial clusters of sites. Our goal is to place sites together in a cluster if they.
